Chapter 12 - Hybrid artificial intelligence models for predicting daily runoff

Abstract

Accurate and reliable prediction of runoff from a catchment is vital for optimal planning and management of water resources. In this chapter, the development of four hybrid artificial intelligence (AI) models by embedding support vector regression (SVR) and multilayer perceptron (MLP) neural network with two metaheuristic algorithms, i.e., whale optimization algorithm (WOA) and gray wolf optimizer (GWO), is discussed. Furthermore, the application of the developed hybrid AI models, i.e., SVR-WOA, SVR-GWO, MLP-WOA, and MLP-GWO, is demonstrated through a case study where daily runoff is predicted for the Naula watershed situated in the upper Ramganga River catchment (RRC) of Uttarakhand State, India. In the case study, the significantly effective input parameters and optimal combinations of the input parameters were identified by applying the gamma test. The predicted daily runoff yielded by the SVR-WOA, SVR-GWO, MLP-GWO, and MLP-WOA models was comparatively evaluated with that predicted by the multiple linear regression (MLR) model using three statistical performance evaluation indicators, i.e., root mean square error (RMSE), Pearson correlation coefficient (PCC), and Willmott index (WI), and through three visual inspection diagrams, i.e., Taylor diagram, line, and scatter plots. The results of comparative evaluation revealed that the better performance of the hybrid AI models over the MLR model. Moreover, the comparative results of runoff forecasts within the hybrid AI models indicated the superior accuracy of the SVR-WOA-5 model (RMSE as 223.046m3/s, PCC as 0.733, and WI as 0.755) compared to other hybrid AI models in predicting the daily runoff in the study watershed.